arXiv subjects
Publications and source records attributed to Jun Li.
Two-step aggregation is introduced to facilitate scalable federated learning (SFL) over passive optical networks (PONs). Results reveal that the SFL keeps the required PON upstream bandwidth constant regardless of the number of involved clients, while bringing ~10% learning accuracy improvement.
Obtaining a high-quality frontal face image from a low-resolution (LR) non-frontal face image is primarily important for many facial analysis applications. However, mainstreams either focus on super-resolving near-frontal LR faces or frontalizing non-frontal high-resolution (HR) faces. It is desirable to perform both tasks seamlessly for daily-life unconstrained face images. In this paper, we present a novel Vivid Face Hallucination Generative Adversarial Network (VividGAN) for simultaneously super-resolving and frontalizing tiny non-frontal face images. VividGAN consists of coarse-level and fine-level Face Hallucination Networks (FHnet) and two discriminators, i.e., Coarse-D and Fine-D. The coarse-level FHnet generates a frontal coarse HR face and then the fine-level FHnet makes use of the facial component appearance prior, i.e., fine-grained facial components, to attain a frontal HR face image with authentic details. In the fine-level FHnet, we also design a facial component-aware module that adopts the facial geometry guidance as clues to accurately align and merge the frontal coarse HR face and prior information. Meanwhile, two-level discriminators are designed to capture both the global outline of a face image as well as detailed facial characteristics. The Coarse-D enforces the coarsely hallucinated faces to be upright and complete while the Fine-D focuses on the fine hallucinated ones for sharper details. Extensive experiments demonstrate that our VividGAN achieves photo-realistic frontal HR faces, reaching superior performance in downstream tasks, i.e., face recognition and expression classification, compared with other state-of-the-art methods.
In most Internet of Things (IoT) networks, edge nodes are commonly used as to relays to cache sensing data generated by IoT sensors as well as provide communication services for data consumers. However, a critical issue of IoT sensing is that data are usually transient, which necessitates temporal updates of caching content items while frequent cache updates could lead to considerable energy cost and challenge the lifetime of IoT sensors. To address this issue, we adopt the Age of Information (AoI) to quantify data freshness and propose an online cache update scheme to obtain an effective tradeoff between the average AoI and energy cost. Specifically, we first develop a characterization of transmission energy consumption at IoT sensors by incorporating a successful transmission condition. Then, we model cache updating as a Markov decision process to minimize average weighted cost with judicious definitions of state, action, and reward. Since user preference towards content items is usually unknown and often temporally evolving, we therefore develop a deep reinforcement learning (DRL) algorithm to enable intelligent cache updates. Through trial-and-error explorations, an effective caching policy can be learned without requiring exact knowledge of content popularity. Simulation results demonstrate the superiority of the proposed framework.
The irregular geometry and high inter-slice variability in computerized tomography (CT) scans of the human pancreas make an accurate segmentation of this crucial organ a challenging task for existing data-driven deep learning methods. To address this problem, we present a novel model-driven stack-based fully convolutional network with a sliding window fusion algorithm for pancreas segmentation, termed MDS-Net. The MDS-Net's cost function includes a data approximation term and a prior knowledge regularization term combined with a stack scheme for capturing and fusing the two-dimensional (2D) and local three-dimensional (3D) context information. Specifically, 3D CT scans are divided into multiple stacks to capture the local spatial context feature. To highlight the importance of single slices, the inter-slice relationships in the stack data are also incorporated in the MDS-Net framework. For implementing this proposed model-driven method, we create a stack-based U-Net architecture and successfully derive its back-propagation procedure for end-to-end training. Furthermore, a sliding window fusion algorithm is utilized to improve the consistency of adjacent CT slices and intra-stack. Finally, extensive quantitative assessments on the NIH Pancreas-CT dataset demonstrated higher pancreatic segmentation accuracy and reliability of MDS-Net compared to other state-of-the-art methods.
Element-wise activation functions play a critical role in deep neural networks via affecting the expressivity power and the learning dynamics. Learning-based activation functions have recently gained increasing attention and success. We propose a new perspective of learnable activation function through formulating them with element-wise attention mechanism. In each network layer, we devise an attention module which learns an element-wise, sign-based attention map for the pre-activation feature map. The attention map scales an element based on its sign. Adding the attention module with a rectified linear unit (ReLU) results in an amplification of positive elements and a suppression of negative ones, both with learned, data-adaptive parameters. We coin the resulting activation function Attention-based Rectified Linear Unit (AReLU). The attention module essentially learns an element-wise residue of the activated part of the input, as ReLU can be viewed as an identity transformation. This makes the network training more resistant to gradient vanishing. The learned attentive activation leads to well-focused activation of relevant regions of a feature map. Through extensive evaluations, we show that AReLU significantly boosts the performance of most mainstream network architectures with only two extra learnable parameters per layer introduced. Notably, AReLU facilitates fast network training under small learning rates, which makes it especially suited in the case of transfer learning and meta learning. Our source code has been released (see https://github.com/densechen/AReLU).
Metal halide perovskites are an important class of emerging semiconductors. Their charge dynamics is poorly understood due to limited knowledge of defect physics and charge recombination mechanisms. Nevertheless, classical ABC and Shockley-Read-Hall (SRH) models are ubiquitously applied to perovskites without considering their validity. Herein, an advanced technique mapping photoluminescence quantum yield (PLQY) as a function of both the excitation pulse energy and repetition frequency is developed and employed to examine the validity of these models. While ABC and SRH fail to explain the charge dynamics in a broad range of conditions, the addition of Auger recombination and trapping to the SRH model enables a quantitative fitting of PLQY maps and low-power PL decay kinetics, and extracting trap concentrations and efficacies. Higher-power PL kinetics requires the inclusion of additional non-linear processes. The PLQY mapping developed herein is suitable for a comprehensive testing of theories and is applicable to any semiconductor.
Machine learning methods based on statistical principles have proven highly successful in dealing with a wide variety of data analysis and analytics tasks. Traditional data models are mostly concerned with independent identically distributed data. The recent success of end-to-end modelling scheme using deep neural networks equipped with effective structures such as convolutional layers or skip connections allows the extension to more sophisticated and structured practical data, such as natural language, images, videos, etc. On the application side, vector fields are an extremely useful type of data in empirical sciences, as well as signal processing, e.g. non-parametric transformations of 3D point clouds using 3D vector fields, the modelling of the fluid flow in earth science, and the modelling of physical fields. This review article is dedicated to recent computational tools of vector fields, including vector data representations, predictive model of spatial data, as well as applications in computer vision, signal processing, and empirical sciences.
Edge nodes (ENs) in Internet of Things commonly serve as gateways to cache sensing data while providing accessing services for data consumers. This paper considers multiple ENs that cache sensing data under the coordination of the cloud. Particularly, each EN can fetch content generated by sensors within its coverage, which can be uploaded to the cloud via fronthaul and then be delivered to other ENs beyond the communication range. However, sensing data are usually transient with time whereas frequent cache updates could lead to considerable energy consumption at sensors and fronthaul traffic loads. Therefore, we adopt age of information to evaluate data freshness and investigate intelligent caching policies to preserve data freshness while reducing cache update costs. Specifically, we model the cache update problem as a cooperative multi-agent Markov decision process with the goal of minimizing the long-term average weighted cost. To efficiently handle the exponentially large number of actions, we devise a novel reinforcement learning approach, which is a discrete multi-agent variant of soft actor-critic (SAC). Furthermore, we generalize the proposed approach into a decentralized control, where each EN can make decisions based on local observations only. Simulation results demonstrate the superior performance of the proposed SAC-based caching schemes.
We aim to provide a computationally cheap yet effective approach for fine-grained image classification (FGIC) in this letter. Unlike previous methods that rely on complex part localization modules, our approach learns fine-grained features by enhancing the semantics of sub-features of a global feature. Specifically, we first achieve the sub-feature semantic by arranging feature channels of a CNN into different groups through channel permutation. Meanwhile, to enhance the discriminability of sub-features, the groups are guided to be activated on object parts with strong discriminability by a weighted combination regularization. Our approach is parameter parsimonious and can be easily integrated into the backbone model as a plug-and-play module for end-to-end training with only image-level supervision. Experiments verified the effectiveness of our approach and validated its comparable performance to the state-of-the-art methods. Code is available at https://github.com/cswluo/SEF
We systematically investigate the electronic structure, magnetism and high-temperature superconductivity (SC) in the multi-layer octagraphene and octagraphite (bulk octagraphene). A tight binding model is used to fit the electronic structures of single-layer, multi-layer octagraphenes and octagraphite. We find that the multi-layer octagraphene and octagraphite follow a simple A-A stacking structure from the energy analysis. The van der Waals interaction induces $t_{\perp}\approx0.25$ eV and the hopping integrals within each layers changes little when the layer number $n$ increases. There is a well Fermi-surface nesting with nesting vector $\mathbf{Q}=(π,π)$ for the single-layer octagraphene at half-filling, which can induce a 2D Néel antiferromagnetic order. With increasing the layer number $n\rightarrow\infty$, the Fermi-surface nesting transforms to 3D with nesting vector $\mathbf{Q}=(π,π,π)$ and shows the system has a 3D Néel antiferromagnetic order. Upon doping, the multi-layer octagraphene and octagraphite can enter a high-temperature $s^{\pm}$ SC driven by spin fluctuation. We evaluate the superconducting transition temperature $T_c$ by using the random-phase approximation (RPA), which yields a high $T_c$ even if the layer number $n\geq$ 3. Our study shows that the multi-layer octagraphene and octagraphite are promising candidates for realizing the high-temperature SC.
Intelligent reflecting surface (IRS) is of low-cost and energy-efficiency and will be a promising technology for the future wireless communications like sixth generation. To address the problem of conventional directional modulation (DM) that Alice only transmits single confidential bit stream (CBS) to Bob with multiple antennas in a line-of-sight channel, IRS is proposed to create friendly multipaths for DM such that two CBSs can be transmitted from Alice to Bob. This will significantly enhance the secrecy rate (SR) of DM. To maximize the SR (Max-SR), a general non-convex optimization problem is formulated with the unit-modulus constraint of IRS phase-shift matrix (PSM), and the general alternating iterative (GAI) algorithm is proposed to jointly obtain the transmit beamforming vectors (TBVs) and PSM by alternately optimizing one and fixing another. To reduce its high complexity, a low-complexity iterative algorithm for Max-SR is proposed by placing the constraint of null-space (NS) on the TBVs, called NS projection (NSP). Here, each CBS is transmitted separately in the NSs of other CBS and AN channels. Simulation results show that the SRs of the proposed GAI and NSP can approximately double that of IRS-based DM with single CBS for massive IRS in the high signal-to-noise ratio region.
Federated learning is a privacy-preserving approach to train a global model at a central server by collaborating with wireless devices, each with its own local training data set. In this paper, we present a compressive sensing approach for federated learning over massive multiple-input multiple-output communication systems in which the central server equipped with a massive antenna array communicates with the wireless devices. One major challenge in system design is to reconstruct local gradient vectors accurately at the central server, which are computed-and-sent from the wireless devices. To overcome this challenge, we first establish a transmission strategy to construct sparse transmitted signals from the local gradient vectors at the devices. We then propose a compressive sensing algorithm enabling the server to iteratively find the linear minimum-mean-square-error (LMMSE) estimate of the transmitted signal by exploiting its sparsity. We also derive an analytical threshold for the residual error at each iteration, to design the stopping criterion of the proposed algorithm. We show that for a sparse transmitted signal, the proposed algorithm requires less computationally complexity than LMMSE. Simulation results demonstrate that the presented approach outperforms conventional linear beamforming approaches and reduces the performance gap between federated learning and centralized learning with perfect reconstruction.
In this paper, we present a Large-Scale and high-diversity general Thermal InfraRed (TIR) Object Tracking Benchmark, called LSOTBTIR, which consists of an evaluation dataset and a training dataset with a total of 1,400 TIR sequences and more than 600K frames. We annotate the bounding box of objects in every frame of all sequences and generate over 730K bounding boxes in total. To the best of our knowledge, LSOTB-TIR is the largest and most diverse TIR object tracking benchmark to date. To evaluate a tracker on different attributes, we define 4 scenario attributes and 12 challenge attributes in the evaluation dataset. By releasing LSOTB-TIR, we encourage the community to develop deep learning based TIR trackers and evaluate them fairly and comprehensively. We evaluate and analyze more than 30 trackers on LSOTB-TIR to provide a series of baselines, and the results show that deep trackers achieve promising performance. Furthermore, we re-train several representative deep trackers on LSOTB-TIR, and their results demonstrate that the proposed training dataset significantly improves the performance of deep TIR trackers. Codes and dataset are available at https://github.com/QiaoLiuHit/LSOTB-TIR.
Computational methods have reshaped the landscape of modern biology. While the biomedical community is increasingly dependent on computational tools, the mechanisms ensuring open data, open software, and reproducibility are variably enforced by academic institutions, funders, and publishers. Publications may present academic software for which essential materials are or become unavailable, such as source code and documentation. Publications that lack such information compromise the role of peer review in evaluating technical strength and scientific contribution. Incomplete ancillary information for an academic software package may bias or limit any subsequent work produced with the tool. We provide eight recommendations across four different domains to improve reproducibility, transparency, and rigor in computational biology - precisely on the main values which should be emphasized in life science curricula. Our recommendations for improving software availability, usability, and archival stability aim to foster a sustainable data science ecosystem in biomedicine and life science research.
We calculate a genus-one FJRW invariant of an LG pair $(W_3=x_1^3+x_2^3+x_3^3, μ_3)$ via two different methods. In the first method, we apply the cosection localization technique to get a genus-one three-spin virtual class explicitly and then calculate the target FJRW invariant via self-intersections of the three-spin virtual class. In the second method, we apply the Mixed Spin P-fields method for the pair and calculate the invariant using the localization formula. This invariant is the building block in establishing the all-genera LG/CY correspondence and its determination enables one to compute the all-genera FJRW invariants for the LG pair.
Designing proper time-dependent control fields for slowly varying the system to the ground state that encodes the problem solution is crucial for adiabatic quantum computation. However, inevitable perturbations in real applications demand us to accelerate the evolution so that the adiabatic errors can be prevented from accumulation. Here, by treating this trade-off task as a multiobjective optimization problem, we propose a gradient-free learning algorithm with pulse smoothing technique to search optimal adiabatic quantum pathways and apply it to the Landau-Zener Hamiltonian and Grover search Hamiltonian. Numerical comparisons with a linear schedule, local adiabatic theorem induced schedule, and gradient-based algorithm searched schedule reveal that the proposed method can achieve significant performance improvements in terms of the adiabatic time and the instantaneous ground-state population maintenance. The proposed method can be used to solve more complex and real adiabatic quantum computation problems.
Gaseous thermal transpiration flows through a rectangular micro-channel are simulated by the direct simulation BGK (DSBGK) method. These flows are rarefied, within the slip and transitional flow regimes, which are beyond many traditional computational fluid dynamic simulation schemes, such as those based on the continuum flow assumption. The flows are very slow and thus many traditional particle simulation methods suffer large statistical noises. The adopted method is a combination of particle and gas kinetic methods and it can simulate micro-flows properly. The simulation results of mass flow rates have excellent agreement with experimental measurements. In another case of 2D channel, the DSBGK comparisons with the DSMC result and the solution of Shakhov equation are also in very good agreement. Another finding from this study is that numerical simulations by including two reservoirs at the channel ends lead to appreciable differences in simulation results of velocity and pressure distributions within the micro-channel. This is due to the inhaling and exhaling effects of reservoirs at the channel ends. Even though excluding those reservoirs may accelerate the simulations significantly by using a single channel in simulations, special attentions are needed because this treatment may over-simplify the problem, and some procedures and results may be questionable. One example is to determine the surface momentum accommodation coefficient by using analytical solution of the mass flow rate obtained in a single-channel problem without the confinement effect of reservoirs at the two ends.
Generative adversarial network (GAN) has attracted increasing attention recently owing to its impressive ability to generate realistic samples with high privacy protection. Without directly interactive with training examples, the generative model can be fully used to estimate the underlying distribution of an original dataset while the discriminative model can examine the quality of the generated samples by comparing the label values with the training examples. However, when GANs are applied on sensitive or private training examples, such as medical or financial records, it is still probable to divulge individuals' sensitive and private information. To mitigate this information leakage and construct a private GAN, in this work we propose a Rényi-differentially private-GAN (RDP-GAN), which achieves differential privacy (DP) in a GAN by carefully adding random noises on the value of the loss function during training. Moreover, we derive the analytical results of the total privacy loss under the subsampling method and cumulated iterations, which show its effectiveness on the privacy budget allocation. In addition, in order to mitigate the negative impact brought by the injecting noise, we enhance the proposed algorithm by adding an adaptive noise tuning step, which will change the volume of added noise according to the testing accuracy. Through extensive experimental results, we verify that the proposed algorithm can achieve a better privacy level while producing high-quality samples compared with a benchmark DP-GAN scheme based on noise perturbation on training gradients.